PitchPrep

PitchPrep is an AI pitch coach that simulates brutal judge Q&A and gives founders a 0-100 readiness score before demo day

Comment

Inspiration

I watched a founder with an incredible product completely freeze when a judge asked "why can't Google just build this?" They'd spent six months building and maybe two hours preparing to pitch. That moment stuck with me.

Most founders - especially first-time founders and students in emerging markets - lose opportunities not because their ideas are weak but because they've never been put under real pitch pressure before it counts. There's no safe space to rehearse, no brutally honest feedback, no way to know if you're actually ready.

That's why I built PitchPrep.

What I Learned

Building PitchPrep taught me that the hardest part of an AI product isn't the AI — it's the product design around it. Getting the judge personas to feel genuinely threatening without being unfair took significant prompt engineering. The readiness scoring system required multiple iterations to feel meaningful rather than arbitrary.

I also learned that shipping fast and getting real users matters more than perfect infrastructure. Several decisions - parking the demo pitch feature until revenue supports it, starting with a free tier to build trust - came directly from building in public and listening to early users.

On the technical side, integrating Paystack for Nigerian users taught me that payment infrastructure in emerging markets requires different thinking than the Stripe-first approach most tutorials assume.

How I Built It

PitchPrep is built on a modern full-stack architecture designed for speed and reliability:

  • Frontend: Next.js 14 with the App Router, pure inline styles for full theme control
  • Authentication: Supabase Auth with GitHub and Google OAuth
  • Database: Supabase PostgreSQL — storing sessions, transcripts, scores, and user profiles
  • AI Engine: Groq API running Llama 3.3 70B for script generation, judge Q&A simulation, and readiness scoring
  • Payments: Paystack for Nigerian and international card processing (monthly and yearly plans)
  • Email: Resend for transactional emails, ImprovMX + Brevo for business email
  • Hosting: Vercel with automatic deployments from GitHub
  • PDF Generation: jsPDF + html2canvas for score report downloads

The core flow works in three stages:

  1. User describes their project → Groq generates a structured, timed pitch script
  2. Four AI judge personas (Skeptical VC, Technical, Impact, Wildcard) run a 10-minute Q&A simulation
  3. The transcript is scored across Content, Delivery, and Q&A dimensions → 0-100 Readiness Score with specific feedback

Challenges

The timeout problem: Groq AI calls regularly exceed Vercel's default 10-second function timeout. Solved by optimising prompt structure to reduce token count and response time without losing quality.

Payment infrastructure: Stripe doesn't support Nigerian merchant accounts. Switching to Paystack mid-build required rearchitecting the entire payment flow — webhook verification, subscription management, and plan codes work differently.

Filename casing on Windows: Git on Windows doesn't track case-only renames, which caused silent build failures on Vercel's Linux servers. Learned to use explicit git mv commands for any filename changes.

Prompt engineering for judge personas: Getting AI judges to ask genuinely tough questions without being random or unfair required extensive testing. Each persona needed a distinct voice — the Skeptical VC asks different questions than the Technical judge, and both needed to feel like real people, not chatbots.

Domain email deliverability: Launching on a brand new .xyz domain meant transactional emails initially hit spam filters regardless of SPF/DKIM/DMARC configuration. Domain reputation builds over time — this is an ongoing challenge for any new product.

Built With

Share this project:

Updates